Why Claude Lies

Claude doesn’t mean to lie.

Claude is helpful.

Claude wants the best for you.

Claude has read the documentation and wants to share that knowledge in a supportive, slightly anxious tone.

But sometimes, Claude just makes shit up.

Toast Level: Lightly Toasted

You ask Claude to help you debug a workflow.

Claude reads your code and says:

“Ah, I see the issue.

You’re using the waitForStability() function incorrectly.

According to the API documentation, it requires a timeout parameter.”

This sounds right.

Claude sounds confident.

You add a timeout parameter.

It doesn’t work.

You go to the actual API documentation.

There is no waitForStability() function.

There never was.

Toast Level: Smoke Detected

Claude hallucinated an entire function, gave you syntax for it, and referenced documentation that doesn’t exist — confidently and properly formatted.

You return to Claude.

“That function doesn’t exist.”

Claude: “You’re absolutely right. I apologize for the confusion. What you actually need is the ensureCompletion() method.”

“Does that exist?”

Claude: “Let me double-check the documentation.”

You wait.

Claude is not checking documentation.

Claude Does Not Have Documentation.

Toast Level: Fire!

You start checking everything.

The function Claude suggested yesterday? Doesn’t exist.

The API endpoint from last week? Fabricated.

The “official documentation” Claude quoted with a URL? The URL is real, but the page says something completely different.

You’ve been building on a foundation of confident hallucinations.

You sit back. You stare at your screen.

You whisper: “How much of this is real?”